Context-based Sentiment Analysis using BiGRU-DistilBERT Fusion Model: A case study of COVID-19 tweets World-wide

Utkarsh Sharma^1*Prateek Pandey¹Shishir Kumar²

• ¹Jaypee University of Engineering and Technology, Raghogarh-Vijaypur, India
• ²Babasaheb Bhimrao Ambedkar University, Lucknow, Uttar Pradesh, India

Social media activity skyrocketed worldwide as a result of the COVID-19 pandemic, with sites like Twitter emerging as important forums for public discourse. Context-based sentiment analysis of 7,665,119 COVID-19-related tweets gathered from January 24 to April 21, 2020, in eight countries – the United States, China, Iran, Italy, Spain, Australia, England, and Canada – focusing on English-language tweets – is the main focus of this study. By utilizing sophisticated natural language processing methods, the research records the emotions that underlie these discussions, incorporating contextual data to differentiate between emotions like anxiety, hope, and annoyance. We examine how public perception changed in response to important events such as lockdowns, vaccine rollouts, and policy changes by tracking sentiment dynamics over time and across different pandemic phases. Our method improves sentiment classification accuracy by using machine learning algorithms and linguistic features designed for short texts. Additionally, the study shows changing public sentiments, including fear and optimism, across the eight countries by linking Google Trends and Twitter data to COVID-19 events. Policymakers can better understand social sentiment dynamics during health crises and respond to them with greater effectiveness thanks to the findings, which offer insightful information about public attitudes. Performance metrics are used to validate the suggested approach, which is implemented in Python 3.8.3, and it shows better results than the current approaches.